def get(self, name, dataset):
model = load_model(name)
metadata = load_model_metadata(name)
df = get_dataset(dataset)
# TODO: make it optional
if metadata['configuration']:
configuration = load_configuration(metadata['dataset'],
metadata['configuration'])
modify(df, configuration)
label = metadata['label']
data_x = df.drop(label, axis=1)
data_y = df[label]
predicts = model.predict_classes(data_x)
score = float((data_y == predicts).sum() / predicts.size)
conf_matrix = confusion_matrix(data_y, predicts)
plots = OrderedDict()
plot_classification_predictions(data_y,
predicts,
orientation='vertical',
stacked=False)
plots['predictions'] = plot_to_base64()
return {
'score': score,
'plots': plots,
'confusionMatrix': conf_matrix.tolist()
}
def get(self, training, testing):
train = get_dataset(training, 5)
test = get_dataset(testing, 5)
result = {
'length': len(train.columns) == len(test.columns),
'columns': False,
'types': False,
}
if result['length']:
result['columns'] = bool((train.columns == test.columns).all())
result['types'] = bool(
(train.dtypes.asobject == test.dtypes.asobject).all())
return result
def post(self):
dataset = api.payload['dataset']
df = get_dataset(dataset)
if 'configuration' in api.payload:
configuration_name = api.payload['configuration']
configuration = load_configuration(dataset, configuration_name)
modify(df, configuration)
df = shuffle(df)
label = api.payload['labelColumn']
data_x = df.drop(label, axis=1)
data_y = df[label]
input_dim = data_x.columns.size
output_dim = data_y.unique().size
epochs = api.payload['epochs']
layers = api.payload['layers']
for layer in layers:
if layer <= 0:
return "Invalid layer", 400
validation_split = api.payload[
'validationSplit'] if 'validationSplit' in api.payload else 0
keras.backend.clear_session()
model = ClassificationModel(input_dim, output_dim, layers)
history = model.fit(x=data_x,
y=data_y,
epochs=epochs,
verbose=2,
validation_split=validation_split)
predicts = model.predict_classes(data_x)
score = float((data_y == predicts).sum() / predicts.size)
conf_matrix = confusion_matrix(data_y, predicts)
plots = OrderedDict()
plot_history_accuracy(history)
plots['accuracy'] = plot_to_base64()
plot_history_loss(history)
plots['loss'] = plot_to_base64()
plot_classification_predictions(data_y,
predicts,
orientation='vertical',
stacked=False)
plots['predictions'] = plot_to_base64()
return {
'score': score,
'plots': plots,
'confusionMatrix': conf_matrix.tolist()
}
def post(self):
dataset = api.payload['dataset']
df = get_dataset(dataset)
if 'configuration' in api.payload:
configuration_name = api.payload['configuration']
configuration = load_configuration(dataset, configuration_name)
modify(df, configuration)
label = api.payload['labelColumn']
data_x = df.drop(label, axis=1)
data_y = df[label]
input_dim = data_x.columns.size
output_dim = data_y.unique().size
epochs = api.payload['epochs']
layers = api.payload['layers']
for layer in layers:
if layer <= 0:
return "Invalid layer", 400
train_scores = []
test_scores = []
kfolds = api.payload['kfolds'] if 'kfolds' in api.payload else 10
kf = StratifiedKFold(n_splits=kfolds, shuffle=True)
for train_indices, test_indices in kf.split(data_x, data_y):
train_x, train_y = data_x.iloc[train_indices], data_y.iloc[
train_indices]
test_x, test_y = data_x.iloc[test_indices], data_y.iloc[
test_indices]
keras.backend.clear_session()
model = ClassificationModel(input_dim, output_dim, layers)
model.fit(x=train_x, y=train_y.values, epochs=epochs, verbose=2)
predicts = model.predict_classes(train_x)
score = float((train_y == predicts).sum() / predicts.size)
train_scores.append(score)
predicts = model.predict_classes(test_x)
score = float((test_y == predicts).sum() / predicts.size)
test_scores.append(score)
plots = OrderedDict()
plot_cross_validation(train_scores, test_scores, plot_type='bar')
plots['crossValidation'] = plot_to_base64()
plot_cross_validation(train_scores, test_scores, plot_type='plot')
plots['crossValidation2'] = plot_to_base64()
return {
'trainScores': train_scores,
'testScores': test_scores,
'plots': plots,
}
def get(self, dataset, rows):
df = get_dataset(dataset, rows, True)
columns = df.columns.values.tolist()
column_types = list(map(lambda t: t.name, df.dtypes.values))
data_preview = df.values.tolist()
return {
'columns': columns,
'columnTypes': column_types,
'rows': data_preview
}
def get(self, dataset, configuration=None):
result = try_load(dataset, configuration)
if result:
return result
df = get_dataset(dataset)
if configuration:
loaded_configuration = load_configuration(dataset, configuration)
modify(df, loaded_configuration)
columns = []
for column_name in df:
series = df[column_name]
describe = series.describe()
is_numeric = np.issubdtype(series.dtype.type, np.number)
if is_numeric:
descriptive_statistics = {
"count": int(describe["count"]),
"mean": describe["mean"],
"std": describe["std"],
"min": describe["min"],
"p25": describe['25%'],
"p50": describe["50%"],
"p75": describe["75%"],
"max": describe["max"]
}
else:
descriptive_statistics = {
"count": int(describe["count"]),
"unique": int(describe["unique"])
}
plots = {}
plot_histogram(series, column_name, is_numeric)
plots['histogram'] = plot_to_base64()
if is_numeric:
plot_box_and_violin(series)
plots['boxplot'] = plot_to_base64()
columns.append({
"name": column_name,
"type": series.dtype.name,
"numeric": is_numeric,
"descriptiveStatistics": descriptive_statistics,
"plots": plots
})
result = {
"columns": columns
}
store(dataset, configuration, result)
return result
def get(self, dataset):
df = get_dataset(dataset)
columns = []
for column in df:
series = df[column]
columns.append({
'name': column,
'type': series.dtype.name,
'hasNA': bool(series.isnull().any())
})
return {'columns': columns}
def post(self):
experiment_name = api.payload['name']
train_dataset_name = api.payload['trainingDataset']
test_dataset_name = api.payload['testingDataset']
label_column_name = api.payload['labelColumnName']
train = get_dataset(train_dataset_name)
test = get_dataset(test_dataset_name)
train_x, train_y_raw = self._split_xy(train, label_column_name)
test_x, test_y_raw = self._split_xy(test, label_column_name)
train_y, classes = train_y_raw.factorize(sort=True)
test_y, _ = test_y_raw.factorize(sort=True)
# prepare model
model = keras.Sequential([
keras.layers.Dense(16,
activation=tf.nn.relu,
input_shape=train_x.shape[1:]),
keras.layers.Dense(len(classes), activation=tf.nn.softmax)
])
model.summary()
# compile model
model.compile(optimizer=tf.train.AdamOptimizer(),
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
# train
model.fit(train_x, train_y, epochs=15, validation_split=0.1)
# test
test_loss, test_acc = model.evaluate(test_x, test_y)
keras.models.save_model(model,
'data/models/{}.h5py'.format(experiment_name))
return {'testLoss': test_loss, 'testAccuracy': test_acc}
def put(self, dataset):
ratio = api.payload['ratio']
shuffle = api.payload['shuffle']
train_dataset_name = api.payload['trainDataset']
test_dataset_name = api.payload['testDataset']
df = get_dataset(dataset)
train, test = train_test_split(df, ratio, shuffle)
save_dataset(train, train_dataset_name)
save_dataset(test, test_dataset_name)
return None, 200
def put(self, dataset):
df = get_dataset(dataset)
columns = api.payload['columns']
if 'newDatasetName' in api.payload and api.payload['newDatasetName']:
dataset_name = api.payload['newDatasetName']
modify(df, columns)
save_dataset(df, dataset_name)
if 'configurationName' in api.payload and api.payload[
'configurationName']:
configuration_name = api.payload['configurationName']
save_configuration(dataset, configuration_name, columns)
return None, 200